The two main constraints that currently limit a broader usage of T cell therapy against viruses are the delay required to obtain specific T cells and the safety of the selection procedure. In the present work we developed a generally applicable strategy that eliminates the need for APC for timing reasons, and the need for infectious viral strains for safety concerns. As a model, we used the selection of T lymphocytes specific for the immunodominant CMV phosphoprotein pp65. PBMC from healthy seropositive donors were first depleted of IL-2R α-chain CD25+ cells and were then stimulated for 24–96 h with previously defined peptide Ags or with autologous PBMC infected with a canarypox viral vector encoding the total pp65 protein (ALVAC-pp65). Subsequent immunomagnetic purification of newly CD25-expressing cells allowed efficient recovery of T lymphocytes specific for the initial stimuli, i.e., for the already known immunodominant epitope corresponding to the peptides used as a model or for newly defined epitopes corresponding to peptides encoded by the transfected pp65 protein. Importantly, we demonstrated that direct PBMC stimulation allowed recovery not only of CD8+ memory T lymphocytes, but also of the CD4+ memory T cells, which are known to be crucial to ensure persistence of adoptively transferred immune memory. Finally, our analysis of pp65-specific T cells led to the identification of several new helper and cytotoxic epitopes. This work thus demonstrates the feasibility of isolating memory T lymphocytes specific for a clinically relevant protein without the need to prepare APC, to use infectious viral strains, or to identify immunodominant epitopes.